Kiwi-Edit: Versatile Video Editing via Instruction and Reference Guidance
This work addresses the challenge of complex visual nuances in video editing for AI and creative professionals, representing an incremental advancement through novel data generation and architecture.
The paper tackles the problem of precise visual control in instruction-based video editing by introducing a scalable data generation pipeline to create high-quality training quadruplets, which leads to a new state-of-the-art model, Kiwi-Edit, that significantly improves instruction following and reference fidelity.
Instruction-based video editing has witnessed rapid progress, yet current methods often struggle with precise visual control, as natural language is inherently limited in describing complex visual nuances. Although reference-guided editing offers a robust solution, its potential is currently bottlenecked by the scarcity of high-quality paired training data. To bridge this gap, we introduce a scalable data generation pipeline that transforms existing video editing pairs into high-fidelity training quadruplets, leveraging image generative models to create synthesized reference scaffolds. Using this pipeline, we construct RefVIE, a large-scale dataset tailored for instruction-reference-following tasks, and establish RefVIE-Bench for comprehensive evaluation. Furthermore, we propose a unified editing architecture, Kiwi-Edit, that synergizes learnable queries and latent visual features for reference semantic guidance. Our model achieves significant gains in instruction following and reference fidelity via a progressive multi-stage training curriculum. Extensive experiments demonstrate that our data and architecture establish a new state-of-the-art in controllable video editing. All datasets, models, and code is released at https://github.com/showlab/Kiwi-Edit.